基于机器学习的元素录井宏观煤岩类型识别方法研究OA

To address the problems of traditional logging methods being easily disturbed by wellbore conditions and poor regional adaptability of models in identifying macroscopic coal rock types,and to provide reliable technical support for coalbed methane reservoir evaluation and sweet spot prediction,taking the coal seams of the Benxi formation in the eastern Ordos basin as the research object,an intelligent identification method combining elemental logging data and machine learning is adopted.Coal seams are accurately identified through density-ash content inverse envelope,the SMOTE algorithm is used to balance the small sample dataset,principal component analysis is combined to reduce the dimensionality of 7 elements including Al,Ca and Fe,three machine learning models(random forest,XGBoost,and support vector machine)are compared to select the optimal model,and blind well verification and gas content correlation analysis are carried out.The research results show that:① The accuracy of coal seam identification by density-ash content inverse envelope reaches 82.9%,and the relative errors of the constructed industrial component calculation models are all less than 27%.② The cumulative contribution rate of the first three principal components extracted by principal component analysis is 83.3%,which can effectively characterize the original element information.③ The XGBoost model has the best identification effect,with a macro-average F1 value of 0.92 on the test set and an average blind well verification accuracy of 89.45%,and the identification accuracy of dull coal reaches 86%.④ There is a positive correlation between coal rock brightness and gas content,with bright coal having the strongest adsorption capacity and higher production potential.It is concluded that the identification method integrating elemental logging and the XGBoost model can effectively reduce environmental interference,improve the accuracy and robustness of macroscopic coal rock type identification,and provide an important technical approach and theoretical basis for high-quality coalbed methane reservoir prediction and productivity evaluation.